Intraday solar energy forecasts are increasingly required for various applications such as electricity trading and microgrids management. Forecasting solar power using meteorological geostationary satellites has proven to be more reliable than numerical weather prediction (NWP) models for a time horizon of up to 4 or 6 hours. This technique has now been operational for a decade, and many commercial weather services include it in their PV production forecasts.

However, current satellite-based solar irradiance forecasts rely on a basic hypothesis: that the cloud cover seen by the satellite presents a horizontal motion (advection) with a constant cloud border shape for a few hours. While this statement is pertinent for a major part of weather situations over Western Europe, the performance of such a satellite-based forecast over a given site is very sensitive to weather stability. In particular, it can strongly decrease in cases of convection, fog, or large depressions.

In this work, we assess the impact of different North Atlantic weather regimes on satellite-based forecast reliability for several sites in Western Europe. The forecasts are generated four hours ahead with a 15-minute time step, using cloud cover extrapolation techniques, and are validated against pyranometer observations. An 8-year backtest is computed, and forecast errors are calculated according to different weather regimes: Atlantic Ridge, Scandinavian Blocking, NAO+, and NAO- in summer and winter. At a 2-hour time horizon, the comparison between the global irradiance forecast and pyranometric observations showed a relative root mean square error (rRMSE) of 35.8% for all winter and summer seasons from 2016 to 2024. The highest rRMSE occurred during the Atlantic Ridge regime (41.5%), while the Scandinavian Blocking regime showed the lowest rRMSE (27.5%). The impact of NAO regimes depends on the period from 2016 to 2024; forecasts are higher during NAO+ occurrences and lower for NAO- occurrences. This trend is reversed after 2020.

These variations in weather regime frequencies directly impact forecast errors, emphasizing the importance of large-scale atmospheric patterns in solar energy forecast reliability. As weather regimes can be predicted several days in advance, this analysis provides useful information to anticipate the magnitude of forecast error and therefore adapt suitable decisions for optimizing PV integration management and electricity trading